Embed Size (px)
Citation preview
Low-Temperature Atomic Layer Deposition of High Purity, Smooth,Low Resistivity Copper Films by Using Amidinate Precursor andHydrogen PlasmaZheng Guo,† Hao Li,‡ Qiang Chen,† Lijun Sang,† Lizhen Yang,† Zhongwei Liu,*,† and Xinwei Wang*,‡
†Laboratory of Plasma Physics and Materials, Beijing Institute of Graphic Communication, Beijing 102600, China‡School of Advanced Materials, Shenzhen Graduate School, Peking University, Shenzhen 518055, China
*S Supporting Information
ABSTRACT: Agglomeration is a critical issue for depositingcopper (Cu) thin films, and therefore, the deposition shouldbe preferably performed below 100 °C. This work explores anatomic layer deposition (ALD) process for copper thin filmsdeposited at temperature as low as 50 °C. The processemploys copper(I)-N,N′-diisopropylacetamidinate precursorand H2 plasma, which are both highly reactive at lowtemperature. The deposition process below 100 °C followsan ideal self-limiting ALD fashion with a saturated growth rateof 0.071 nm/cycle. Benefitting from the low processtemperature, the agglomeration of Cu thin films is largelysuppressed, and the Cu films deposited at 50 °C are pure,continuous, smooth, and highly conformal, with the resistivitycomparable to PVD Cu films. In-situ reaction mechanism studies by using quartz crystal microbalance and optical emissionspectroscopy are followed, and the results confirm the high reactivity of the Cu amidinate precursor at low temperature. To thebest of our knowledge, this is the first successful implementation of metal amidinate precursors for low-temperature (∼50 °C)ALD process. The strategy of using metal amidinate precursors in combination with highly reactive H2 plasma is believed to beextendable for the depositions of other pure metals at low temperature.
1. INTRODUCTION
Due to the low electrical resistivity and high electromigrationresistance, copper (Cu) has been employed as the primaryinterconnect material in microelectronic devices. As the devicesize continues to scale down, as well as the quick developmentof 3D devices, there is an urgent need on fabrication methodsto deposit high-quality Cu thin films conformally inside narrowtrenches or on sophisticated 3D structures.1 Atomic layerdeposition (ALD) is well known for depositing uniform filmsconformally on high aspect ratio structures, so it is consideredas one of the most important techniques for realizing nanoscaledevice fabrication.2−4
Although many excellent ALD processes have beendeveloped for various materials,3 the existing ALD processesfor Cu metal are not yet completely satisfactory. One criticalproblem with Cu is its low onset temperature for agglomer-ation. It has been suggested5 that the deposition temperatureshould be lower than 100 °C to avoid the agglomeration, inorder to obtain continuous, smooth, thin Cu films. However, asrecently surveyed by Barry et al.,6 most of the reported Cuprocesses were still performed above 100 °C. Examples withthe deposition temperature close to 100 °C include Cu-(dmap)2/ZnEt2 (dmap = dimethylamino-2-propoxide) at 100−120 °C,7 Cu(pyrim)2/ZnEt2 (pyrim = N-ethyl-2-pyrrolylaldi-
minate) at 120−150 °C,8 Cu(dmamb)2/HCOOH/N2H4
(dmamb = dimethylamino-2-methyl-2-butoxide) at 100−170°C,5 and Cu(dmap)2/(HCOOH)/BH3NHMe2 at 130−160°C.9 Besides the deposition temperature issue, these processesalso suffered from various other issues, such as Zn impurityincorporation if using ZnEt2,
8 careful handling required forhydrazine, and catalytic Pd substrate is needed if usingBH3NHMe2.
9 Another combination of Cu(hfac)2/pyridine/H2 (hfac = 1,1,1,5,5,5-hexafluoroacetylacetonate) was reportedto deposit Cu in the temperature range from 100 °C down toroom temperature;10 however, the deposited films tend tocontain higher than normal impurities of carbon and oxygen.In general, the difficulty to achieve low-temperature Cu
deposition is mainly due to the lack of suitable precursors thathave enough reactivity at low temperature and can proceed thesurface reactions in a clean and complete fashion. Notice thatthe reactivity should be sufficient for both the Cu precursor andthe coreactant agent. As for the coreactant agent, an often-usedstrategy is to use energetic plasma to enhance its reactivity,which is also known as plasma-enhanced ALD (PEALD).
Received: June 6, 2015Revised: August 13, 2015Published: August 14, 2015
Article
pubs.acs.org/cm
© 2015 American Chemical Society 5988 DOI: 10.1021/acs.chemmater.5b02137Chem. Mater. 2015, 27, 5988−5996
Dow
nloa
ded
by U
NIV
TO
WN
OF
SHE
NZ
HE
N o
n Se
ptem
ber
9, 2
015
| http
://pu
bs.a
cs.o
rg
Pub
licat
ion
Dat
e (W
eb):
Aug
ust 2
0, 2
015
| doi
: 10.
1021
/acs
.che
mm
ater
.5b0
2137
Several successful implementations of H2 plasma for low-temperature ALD processes have been reported with the Cuprecursor of Cu(acac)2 (acac = acetylacetonate) at 85−140°C,11,12 Cu(thd)2 (thd = 2,2,6,6-tetramethylheptane-3,5-dionate) at 60−180 °C,13 and Cu(dmamb)2 at 100−180°C,14 respectively. H2 plasma is supposed to generate highlyreactive hydrogen atoms that can hydrogenate and eliminatethe organic ligands on surface.15,16 However, not all thehydrogenated products are volatile, so possibly a few percent ofcarbon and oxygen may still leave in the deposited films as theimpurities.14 On the other hand, the atomic hydrogen does notstick very long on Cu surface;17 they may quickly recombineinto H2 molecules and leave the Cu surface when the plasma isturned off. So H2 plasma is unlikely to directly facilitate thefollowing chemisorption step of Cu precursor on freshlyformed Cu surface, and as a result, the growth rate may still below (e.g., only ∼0.02 nm/cycle for Cu(acac)2
11,12 andCu(thd)2
13). Therefore, a low-temperature reactive Cuprecursor is still largely in need to be used along with the H2plasma.In the process of seeking suitable precursors, metal amidinate
compounds have aroused our special attention. Metalamidinates were proposed by Gordon et al.18 as the ALDprecursors to deposit pure metal films by reacting withmolecular H2, and many transition metals (including Cu)have been successfully deposited from the related amidinateprecursors. The Cu films deposited from copper(I)-N,N′-di-sec-butylacetamidinate were quite pure,19 which suggested afavorable clean surface reaction. However, the minimumdeposition temperature (>185 °C18) was still quite high forCu, so the resulting Cu films showed considerable agglomer-ation on SiO2.
19 Also problematic was its slow saturationbehavior,19 which seemed to indicate some kinetic barriers for
this precursor. Therefore, Cu amidinates were not thought tobe applicable for <100 °C ALD process before. But recently,some mechanism studies performed by Dai et al.20 and Ma etal.21−23 in ultrahigh vacuum (UHV) systems revealed asomewhat different story. Ma et al. found23 that thecopper(I)-N,N′-di-sec-butylacetamidinate molecule was actuallyquite reactive at low temperature; it could spontaneouslydissociate and chemisorb on clean Cu surface at −25 °C, andthe actual rate limiting step for the deposition was thedissociation of molecular H2 into H atoms on Cu surface, as theH atoms were the critical species for hydrogenating the surfaceorganic ligands. This finding is very important and intriguing. Itinspires us to think whether the amidinate precursor can beused along with H2 plasma at low temperature, since the H2plasma can provide a rich amount of H atoms so the ratelimiting step may be much accelerated.To implement this idea, we developed a low-temperature Cu
ALD process in this work, by employing copper(I)-N,N′-diisopropylacetamidinate as the Cu source and H2 plasma asthe reducing agent. As speculated, due to the excellent low-temperature reactivity of both species, the deposition processproceeded in an ideal ALD fashion with a reasonably highsaturated growth rate (0.071 nm/cycle) even at temperature aslow as 50 °C. The films deposited at 50 °C were pure,continuous, and smooth, with the resistivity comparable toPVD Cu films. The Cu films could also be conformallydeposited into 10:1 aspect ratio trenches, demonstrating thepromising applications of this process for 3D structuralmicroelectronics. In-situ quartz crystal microbalance and opticalemission spectroscopy measurements were followed to confirmthe high reactivity at low temperature for both reactants. To thebest of our knowledge, this is the first report of implementingmetal amidinate precursors for low-temperature (∼50 °C)
Figure 1. (a) Schematic of the tubular ALD reactor. (b) Schematic layout of the plasma-enhanced ALD cycle for Cu deposition.
Chemistry of Materials Article
DOI: 10.1021/acs.chemmater.5b02137Chem. Mater. 2015, 27, 5988−5996
5989
Dow
nloa
ded
by U
NIV
TO
WN
OF
SHE
NZ
HE
N o
n Se
ptem
ber
9, 2
015
| http
://pu
bs.a
cs.o
rg
Pub
licat
ion
Dat
e (W
eb):
Aug
ust 2
0, 2
015
| doi
: 10.
1021
/acs
.che
mm
ater
.5b0
2137
ALD. As metal amidinates generally show clean surfacechemistry, we believe that our strategy can be extended toother amidinate based precursors, so that more types of high-quality metals can be deposited at a temperature close to roomtemperature.
2. EXPERIMENTAL SECTIONALD Reactor Setup. The deposition was performed in a
homemade tubular ALD reactor, which is schematically illustrated inFigure 1a. The deposition chamber was made of a 60 cm long quartztube, the upstream part of which was wrapped with copper coil.Through this coil, pulsed radiofrequency (RF) power of 13.56 MHzranging from 40−150 W was supplied to generate plasma inside theupstream zone of the quartz tube. A 40 cm long sample holder wasplaced in the center zone of the tube, and sample substrates werenormally placed near the center of the holder for deposition.Deposition Details. Copper(I)-N,N′-diisopropylacetamidinate
(Cu(amd)) was used as the Cu precursor. The Cu(amd) wassynthesized by a literature method.24 1H NMR and thermalgravimetric analysis (Supporting Information Figure S1) were usedto confirm the product and its purity. For deposition experiments, theCu(amd) precursor was kept in a glass container inside an oven at 90°C, and the precursor vapor was delivered into the deposition chamberwith the assist of purified N2 gas (50 sccm) as the carrier gas. As willbe shown in the following, Cu(amd), along with hydrogen plasma, canproduce very good Cu films in an ideal self-limiting ALD fashion, butCu(amd) and H2 gas do not directly react at our low depositiontemperatures (≤100 °C). Thus, in this case, we can safely use pure H2gas (50 sccm) as the purging gas for removing excess Cu precursor.Using H2 gas for purging can also provide particular convenience withour experimental setup, as in the following plasma steps, one onlyneeds to turn on the RF power to generate the hydrogen plasma, butno need to wait for the flow to stabilize as required for switching gastypes. A schematic of the ALD cycles is shown in Figure 1b. Lengths ofthe precursor pulse, purge, and plasma pulse as well as the plasma RFpower were varied to determine the saturation growth conditions. Wefound that (vide infra) a cyclic deposition schedule of 5, 10, 10, and 10
s for the Cu(amd) pulse, purge, plasma pulse, and purge, respectively,with the RF power of 80 W gave out the ideal ALD saturation growthbehavior. Thus, unless otherwise specified, we chose to use this set ofparameters for the later depositions. Silicon wafers and glass slideswere used as the substrates, and no appreciable difference wasobserved for the films deposited on these two types of substrates. Thesubstrate samples were sequentially cleaned by acetone, methanol, andisopropanol and then pretreated by 3 min of H2 plasma prior to theCu deposition. Native oxide on silicon was not intentionally removed.
Film Characterizations. To measure the film thickness, thedeposited Cu films were first scratched and then measured the stepprofiles (Veeco, Dektak 150) across the scratched lines. The obtainedthickness values were also verified by X-ray reflectivity (Bruker, D8Advance) for smooth films and by cross-sectional scanning electronmicroscopy (SEM) (Hitachi, SU8020) for thick films. The surfacemorphology of the films was examined by SEM and atomic forcemicroscopy (AFM) (Veeco, diInnova). The film microstructure wasexamined by transmission electron microscopy (TEM) (JEOL,JEM2100). The composition of the films was analyzed by X-rayphotoelectron spectroscopy (XPS) (Thermo Scientific, K-Alpha). Thesheet resistance of the films was measured by four-point probesmethod (RTS-8).
In-Situ Diagnostics. Quartz crystal microbalance (QCM) andoptical emission spectrometer (OES) were used to in situ investigatethe reaction mechanism. The QCM consisted of a gold-covered AT-cut quartz crystal with an oscillation frequency of ∼6 MHz, and thechange of the frequency was monitored by the Inficon SQC-310controller. The OES spectra were collected by the Acton SepctroProSP-2500 spectrometer, which had a nominal spectral resolution of 0.05nm from 200−900 nm. Photomultiplier tube type detector(Hamamatsu) was used for the time-resolved measurements, inwhich the transient emission signals were acquired by an oscilloscope(Tektronix DP04104) with a temporal resolution of 2 ms.
3. RESULTS
The deposition behavior of using Cu(amd) as the coppersource and hydrogen plasma as the reducing agent was carefully
Figure 2. Self-limiting growth behavior is demonstrated at both 50 and 100 °C. (a) and (b) show how the growth rate approach a constant (a) as thepulse length of the Cu precursor increases, and (b) as the pulse length of the plasma increases, respectively. (c) Growth rate as a function of the RFpower of the plasma. (d) Growth rate as a function of the purge length after each precursor pulse. (e) Film thickness as a function of the number ofALD cycles; the dashed lines represent the corresponding linear fittings. (f) Growth rate as a function of the deposition temperature. Unlessotherwise labeled on the corresponding horizontal axis, the deposition conditions were 5 s, 10 s, 10 s, and 80 W for the Cu(amd) pulse, purge,plasma pulse, and RF power, respectively.
Chemistry of Materials Article
DOI: 10.1021/acs.chemmater.5b02137Chem. Mater. 2015, 27, 5988−5996
5990
Dow
nloa
ded
by U
NIV
TO
WN
OF
SHE
NZ
HE
N o
n Se
ptem
ber
9, 2
015
| http
://pu
bs.a
cs.o
rg
Pub
licat
ion
Dat
e (W
eb):
Aug
ust 2
0, 2
015
| doi
: 10.
1021
/acs
.che
mm
ater
.5b0
2137
studied in the temperature range of 50−140 °C. ExcellentALD-behavior growth was achieved from 50 to 100 °C. TypicalALD-behavior growth curves corresponding to the depositiontemperatures of 50 and 100 °C, are shown in Figure 2a−e.The growth rate versus the pulse length of Cu(amd) typically
saturated in several seconds, as shown in Figure 2a. However,this saturation length was, in fact, not well-defined because ourCu precursor remained in solid phase during the deposition sothe evaporation rate of the precursor gradually decreased as thesmall-size precursor crystallites evaporated first and left thelarge-size crystallites, which could be much slower to evaporate.Consequently, the pulse length required to saturate the growthrate would gradually increase over time. Therefore, the growthrate was always carefully monitored, and excessively long pulseswere used to ensure the saturation. On the other hand, thegrowth rate versus the pulse length of plasma saturateduniformly in about 1 s (Figure 2b), and this value was alsoconsistent with the QCM and OES measurements as will beshown later. The density of the reactive species in the hydrogenplasma was controlled by the supplied RF power. As shown inFigure 2c, as long as enough power was supplied (∼60 W), thegrowth rate did not vary significantly with the RF power. But, ifthe supplied power was too high, the large amount of highlyreactive plasma species could transfer too much energy to thesubstrate, and therefore raise the substrate temperature, whichcould considerably accelerate the unwanted agglomeration ofthe deposited copper thin films (Supporting Information FigureS2). Thus, to keep the agglomeration minimum, we chose touse the RF power of 80 W in the following experiments. Thepurge length after the Cu precursor pulse was also varied, andthe results showed that 10 s of inert gas (H2) purging wassufficient to remove the excess precursor (Figure 2d). Based onthe above results and analysis, we concluded that an excellentALD saturation growth behavior could be obtained withsuitable deposition conditions. A suitable set of the depositionconditions are, for instance, 5 s, 10 s, 10 s, and 80 W for theCu(amd) pulse, purge, plasma pulse, and RF power,
respectively. Unless otherwise specified, we chose to use thisset of parameters for the following part of this study.With the saturation growth conditions, good linear growth
behavior was obtained as shown in Figure 2e, and no obviousincubation cycles were observed in the initial growth. Thetemperature dependence of the growth rate was alsoinvestigated. As shown in Figure 2f, the growth rate remainedfairly constant at around 0.071 nm/cycle below 100 °C, butincreased gradually to 0.095 nm/cycle when the depositiontemperature raised to 140 °C. The increase of the growth ratemight be due to the partial CVD reaction with H2 at theelevated temperature.25 But a more critical issue associated withthe elevated temperature is the accelerated agglomeration forthe deposited thin copper films. The deposition of thin copperfilms is known to suffer from easy agglomeration,19 as thecopper metal has rather poor adhesion to most oxide substratesand the surface copper atoms are quite mobile at an elevatedtemperature.26,27 This temperature effect was also observed inour experiments. As shown in Figure 3, where the surfacemorphology and roughness of the 300-cycle copper filmsdeposited at various temperatures were compared, continuousfilms were only obtained for the depositions performed at 50°C; at 80 °C, the films were barely continuous, and ≥100 °Cclearly discontinuous island-like films were formed. Also, thegrain size (or island size) increased along with the depositiontemperature, which again suggested that the depositiontemperature was the critical factor for agglomeration. Noticethat the obtained upper limit of the deposition temperature toform continuous films in this study was appreciably lower thanthe upper limit suggested previously5 (more than 20 °C lower).This is probably because we used the energetic plasma whichcould locally heat up the substrate and enhance the mobility ofsurface copper atoms during the deposition. In fact, we foundthat increasing the RF power could also result in larger grainsize (Supporting Information Figure S2). Nevertheless, thefilms deposited at 50 °C looked quite good; the films werecontinuous and fairly smooth with an rms roughness of only
Figure 3. Comparison of the SEM top-view images of the 300-cycle Cu films deposited at (a) 50 °C, (b) 80 °C, (c) 100 °C, (d) 120 °C, and (e) 140°C, respectively. (f) Plot of the AFM-measured rms roughness values of these films (see Supporting Information Figure S3 for associated AFMimages).
Chemistry of Materials Article
DOI: 10.1021/acs.chemmater.5b02137Chem. Mater. 2015, 27, 5988−5996
5991
Dow
nloa
ded
by U
NIV
TO
WN
OF
SHE
NZ
HE
N o
n Se
ptem
ber
9, 2
015
| http
://pu
bs.a
cs.o
rg
Pub
licat
ion
Dat
e (W
eb):
Aug
ust 2
0, 2
015
| doi
: 10.
1021
/acs
.che
mm
ater
.5b0
2137
0.72 nm as shown in Figure 3f. Therefore, we focused on thefilms deposited at this temperature for the following character-izations.The quality of the Cu films deposited at 50 °C was carefully
evaluated (Figure 4). The microstructures of the depositedfilms were examined by TEM. As the TEM image and theassociated electron diffraction pattern shown in Figure 4a−b,the deposited Cu films were continuous and polycrystallinewith the grain size around 20 nm. The film purity was examinedby XPS. After 30 s of Ar+ sputtering for removing the surfaceadventitious carbon, no oxygen, nitrogen, or carbon peaks wereobserved in the XPS spectrum (Figure 4d), indicating the highpurity of the deposited Cu films. The film step coverage wasevaluated by depositing the Cu film on a trench-structuredsample with the aspect ratio of 10:1 for the trenches. As thecross-sectional SEM shown in Figure 4c shows, the Cu filmdeposited inside the trench was highly conformal, and uniformfilm thickness was achieved along the entire trench, whichdemonstrated an excellent step coverage of this ALD process.Large-scale deposition uniformity was evaluated by examiningthe thickness of films deposited at various places along thereactor tube. As shown in Figure 4e, the film thickness wasquite uniform along over 30 cm of the deposition zone,demonstrating a fairly good large-scale deposition uniformity.The resistivity was also measured for the films with variousthicknesses. As shown in Figure 4f, the resistivities were allbelow 20 μΩ cm for the Cu films thicker than 15 nm(compared to the bulk Cu resistivity of 1.7 μΩ cm).Particularly, for the ∼33 nm film, the resistivity was only∼5.6 μΩ cm, which was similar to that for a sputtered Cu filmwith the same thickness (5−6 μΩ cm28), indicating that ourdeposited Cu films had a high purity and low surface roughness,because otherwise the film resistivity would be considerably
higher. In addition, the deposited Cu films also passed theScotch tape test, demonstrating very good adhesion.From all the aspects of properties we evaluated, the Cu films
prepared by our ALD process at 50 °C exhibited very highquality, demonstrating the high promise of this process. On theother hand, this process has an intriguingly uncommon featurethat the deposition temperature was appreciably lower than theprecursor temperature (50 °C vs 90 °C). Normally, this wouldresult in the precursor condensation effect, where a largeamount of precursor could condense on the substrate at lowertemperature and become difficult to remove during the purging.To further investigate this regard, as well as to understand themechanism of the surface reactions involved in this ALDprocess, we performed the in situ measurements of QCM andOES as the following.QCM has been frequently used to monitor the growth
process of ALD because the oscillation frequency of the quartzcrystal is highly sensitive to the mass change. However, theoscillation frequency also depends on temperature, and thus,sometimes, spike-like artifacts due to the sudden temperaturechanges during pulsing could occur.29 The situation becomeseven more pronounced and complicated when plasma is used,because the charged energetic plasma could not only transferheat but also interfere with the oscillation circuits.30 Therefore,special cares were taken for the QCM measurements as well asfor the data interpretation. Though our goal was to investigatethe ALD process with the cyclic deposition schedule of 5, 10,10, and 10 s for the Cu(amd) pulse, purge, plasma pulse, andpurge, respectively, we first started with the dummy cycles (i.e.,without dosing Cu(amd)) until the acquired QCM datareached stable (no baseline drifting), and then followed bythe normal deposition cycles. This arrangement allowed theQCM to stabilize at a new temperature due to the heating effect
Figure 4. Characterizations of the ∼20 nm Cu films deposited at 50 °C, including (a) TEM image and (b) electron diffraction pattern showing thefilm microstructure; (c) cross-sectional SEM image showing the conformality of the deposition inside a 10:1 trench; (d) XPS spectrum showing thefilm purity; (e) distribution of the film thickness along the reactor tube; and (f) film resistivity as a function of the film thickness.
Chemistry of Materials Article
DOI: 10.1021/acs.chemmater.5b02137Chem. Mater. 2015, 27, 5988−5996
5992
Dow
nloa
ded
by U
NIV
TO
WN
OF
SHE
NZ
HE
N o
n Se
ptem
ber
9, 2
015
| http
://pu
bs.a
cs.o
rg
Pub
licat
ion
Dat
e (W
eb):
Aug
ust 2
0, 2
015
| doi
: 10.
1021
/acs
.che
mm
ater
.5b0
2137
from the intermittent plasma. A representative set of theacquired QCM data are plotted in Figure 5a, where the last five
dummy cycles prior to the deposition were also included forcomparison. The mass gain at the end of each ALD cycleshowed a clear linear increase with the cycle number. However,the shape of the curve during each ALD cycle was ratherconvoluted by the plasma effect. For a better comparison,enlarged curves of a typical deposition cycle and a typicaldummy cycle are plotted in Figure 5b and c, respectively. Theconcurrences of the dip shape during the plasma-on stage andthe bump shape after the plasma-on stage in both plotsindicated that these shapes were the results of the plasma effect.Therefore, to remove the plasma effect, we took the curveshape from the dummy cycles as the background, andsubtracted it from the original QCM data. The processedQCM curve is shown in Figure 5d, with a typical depositioncycle enlarged in Figure 5e. As one can see, the artifactsinduced by the plasma was successfully eliminated.The curve shown in Figure 5e exhibits many intriguing
features. During the first 5 s when the Cu precursor wassupplied, the mass gain of QCM continued increasing to m3,suggesting that the chemisorption and possibly physisorption ofthe Cu(amd) precursor occurred on the surface. Then, themass gain dropped gradually to m2 during the following 10 s of
purging, suggesting that the some adsorbed species could bepurged away. The purgeable species were likely those weaklyphysisorbed precursor dimer molecules and also possibly fromthe recombination of the chemisorbed precursor monomers.Next, when the plasma was turned on, a steep drop of the massgain was observed in the initial ∼1 s, and then the mass gainremained almost constant until the end of the ALD cycle (m1).The drop of the mass gain was probably due to thehydrogenation reaction that removed the organic ligands onsurface and left only Cu atoms, which had a smaller mass. Thisreaction probably completed in about 1 s, as the steep drop inmass gain lasted only for that period of time. To facilitate thelater discussion on the reaction mechanism, we extracted them1, m2, and m3 for each ALD cycle, and plotted the m1 and theratios of m2/m1 and m3/m1 as shown in Figure 5f. Except forthe first cycle which was strongly affected by the substratesurface, for the following cycles, the ratios of m2/m1 and m3/m1remained fairly constant at 2.2 and 5.7, respectively. Noticingthat the molecular weight ratio of Cu(amd) and Cu is 3.22, theratio of m2/m1, which fell between 1 and 3.22, indicated thatsome dissociative chemisorption must occur along with theproduction of some volatile species. In fact, the detailedmechanism involved in the surface reaction is quite complex,and indeed it is not possible to accurately obtain the wholepicture solely by the QCM measurements, but, thanks to Ma etal.22,23 who have carefully studied a similar compound(copper(I)-N,N′-di-sec-butylacetamidinate) on Cu surface, wecan comparatively analyze our results with their results anddraw insightful conclusions (see Discussion). In addition, wemay assume that (m3 − m2)/m1 (i.e., 3.5 in this case) iscorrelated with the layers of the physisorbed Cu(amd)molecules (possibly sit in the dimeric form outside thechemisorbed layer) on each deposited Cu atom, and accordingto our data, only about one such layer was expected, indicatingthe absence of severe precursor condensation with ourexperimental conditions.OES is another useful tool to study a plasma system, as many
excited gas species in plasma have their own characteristicemission wavelengths. However, the emission spectrumcollected during our deposition (Supporting InformationFigure S4) was overwhelmed by the strong emissions fromatomic H* (e.g., Hα at 656.3 nm) and molecular H2* (e.g.,Fulcher-α d3Π u → a3Σ g
+ bands at 590−640 nm,31 G1Σ g+ →
B1Σ u+ bands at 453−464 nm,32 etc.). The emissions from the
expected hydrocarbon/hydroaminocarbon byproducts wereprobably too weak to be observed, so the time-resolved OESspectra were only taken for the hydrogen species, that is, H*and H2*. The OES results (Supporting Information Figures S4and S5) suggested that an appreciable amount of atomichydrogen was consumed during the first ∼1 s when the plasmawas on and the consumption of atomic hydrogen depended onthe precursor pulse length and saturated at 5 s, whichmanifested again the self-limiting growth behavior as indicatedby Figure 2a.
4. DISCUSSIONAchieving low-temperature Cu ALD generally requires twoconditions: (1) the Cu precursor needs to be reactive enough atlow temperature to be chemisorbed on the substrate surface, sothat an ALD cycle can be initiated; (2) in the second half of anALD cycle, the organic component of the chemisorbed speciesneeds to be removed by a clean and efficient way, so that thenext ALD cycle can proceed. In this work, we employed highly
Figure 5. QCM measurements during ALD of Cu at 50 °C. (a) Curveof the acquired mass gain versus the deposition time for five dummycycles followed by 10 normal ALD cycles. Enlarged curves of twoboxed regions in (a) for a normal cycle and a dummy cycle are plottedin (b) and (c), respectively. By removing the plasma effect, the massgain curves shown in (a) and (b) are updated as (d) and (e),respectively. (f) Plot of the cyclic mass gain of m1 (green full circles)and the ratios of m2/m1 (red open circles) and m3/m1 (red opendiamonds), where the definitions of m1, m2, and m3 are illustrated in(e).
Chemistry of Materials Article
DOI: 10.1021/acs.chemmater.5b02137Chem. Mater. 2015, 27, 5988−5996
5993
Dow
nloa
ded
by U
NIV
TO
WN
OF
SHE
NZ
HE
N o
n Se
ptem
ber
9, 2
015
| http
://pu
bs.a
cs.o
rg
Pub
licat
ion
Dat
e (W
eb):
Aug
ust 2
0, 2
015
| doi
: 10.
1021
/acs
.che
mm
ater
.5b0
2137
reactive Cu(amd) precursor along with H2 plasma to fulfill boththe requirements. As the results shown previously, benefitedfrom the low process temperature of 50 °C that minimized theagglomeration, very high quality Cu films were obtained. Inparticular, the deposited films were fairly smooth, continuous,low resistive, and highly conformal, demonstrating the highpromise of this process for conformal Cu thin film depositionon 3D structures. On the other hand, the surface chemistryinvolved in the deposition process must be favorable forobtaining the high quality films. Therefore, we discuss on thisregard in the following.In general, metal amidinates are an interesting class of ALD
precursors, as they have quite a few unique features that areconsidered preferable for depositing pure metals. For instance,the amidinate ligands contain only the elements of C, N, and H,so there is no need for concern about the oxygen incorporation.This is particularly important for a low-temperature plasma-assisted process, because oxygen-containing moieties are likelyto be hydrogenated by the plasma to form alcohols and water,which are less volatile than alkanes/amines/ammonia andtherefore difficult to be completely purged away at lowtemperature. On the other hand, the bidentate amidinateligands also provide enough stability to the coordinatedcompound, so the Cu(amd) molecule is thermally quite stablebelow 100 °C. This is also very important because it guaranteesthat, during the deposition at 50 °C, the physisorbed Cu(amd)molecules do not self-decompose and therefore can be removedas a whole during the following purging steps.As for the dissociative chemisorption process, Ma et al.22,23
have carefully studied a similar Cu amidinate compound (i.e.,copper(I)-N,N′-di-sec-butylacetamidinate, denoted as Cu-(Buamd) in the following) by combined the experiments oftemperature-programmed desorption and X-ray photoelectronspectroscopy in an UHV system, and revealed that theCu(Buamd) can be dissociatively chemisorbed on fresh Cusurface and partially decompose to form sec-butyl and N-sec-butylacetamidinate moieties, with a portion of the latterdesorbed from the surface at room temperature. We expectthat the surface chemistry involved for our Cu(amd) precursor(i.e., copper(I)-N,N′-diisopropylacetamidinate) should besimilar. During the deposition processed at 50 °C, thechemisorbed Cu(amd) molecules are expected to partiallydecompose to form isopropyl and N-isopropylacetamidinatemoieties on Cu surface. If we further assume that some surfaceN-isopropylacetamidinate moieties can desorb from the surfaceas well, then the desorbed portion can be estimated from them2/m1 ratio from the QCM results. Based on our data, the
desorbed portion of the N-isopropylacetamidinate moieties wasestimated to be 65% (counted based on the total amount of theoriginal N,N′-diisopropylacetamidinate moieties). This valuewas fairly high, and it indicated a favorable surface chemistry,because, otherwise, too many bulky amidinate moieties onsurface would block the surface reaction sites and result in lowdeposition rate. In fact, the saturated ALD growth rate in ourcase was ∼0.071 nm/cycle, which corresponded to about 6 Cuatoms deposited per 1 nm2 surface area for each ALD cycle.Considering the fact that the surface N,N′-diisopropylacetami-dinate and N-isopropylacetamidinate moieties occupy thesurface area of roughly 1.0 nm × 0.4 nm and 0.6 nm × 0.4nm, respectively, it would be impossible to accommodate sixamidinate surface moieties within 1 nm2 area. Therefore, it isfairly reasonable to expect a large portion of the decomposedN-isopropylacetamidinate moieties to release from the surfacein the chemisorption step. On the other hand, the unreleasedN-isopropylacetamidinate moieties were unlikely to furtherdecompose at our low process temperature of 50 °C.After the above complex decomposition and desorption
process, a compact chemisorption layer consisting of theundecomposed N,N′-diisopropylacetamidinate, isopropyl, andunreleased N-isopropylacetamidinate surface moieties waslikely formed. This layer was compact enough to preventadditional Cu(amd) molecules from approaching the Cusurface, and therefore, further surface chemistry reaction wasstopped. Although additional Cu(amd) molecules might stillcontinue to pile up (probably in the dimeric form) on top ofthe chemisorption layer and form a physisorption layer, asmanifested as the difference of m3 − m2 from the QCM results,this physisorbed layer could be easily removed in the followingpurging step, so the self-limiting behavior was still perfectlypreserved.As shown in previous XPS results (Figure 2d), the deposited
Cu films contained very low impurity level, which suggestedthat all the surface moieties in the chemisorption layer could becompletely hydrogenated in the second half of an ALD cycle,provided that rich amount of atomic hydrogen was suppliedfrom the H2 plasma. The hydrogenated products were probablypropane, N-isopropylacetamidine, and N,N′-diisopropylaceta-midine, which were all fairly volatile and could be completelyremoved by sufficient purging.As a summary, the proposed ALD reaction mechanism is
schematically depicted in Figure 6. The involved surfacereactions are regarded as clean reactions, as they allow theentire process to proceed in an ideal ALD fashion and producehigh-quality thin films.
Figure 6. Schematic depiction of the proposed ALD reaction mechanism, where the dash arrows are used to denote partial conversions.
Chemistry of Materials Article
DOI: 10.1021/acs.chemmater.5b02137Chem. Mater. 2015, 27, 5988−5996
5994
Dow
nloa
ded
by U
NIV
TO
WN
OF
SHE
NZ
HE
N o
n Se
ptem
ber
9, 2
015
| http
://pu
bs.a
cs.o
rg
Pub
licat
ion
Dat
e (W
eb):
Aug
ust 2
0, 2
015
| doi
: 10.
1021
/acs
.che
mm
ater
.5b0
2137
5. CONCLUSIONSIn this work, we developed a low-temperature Cu ALD processby employing the highly reactive copper(I)-N,N′-diisopropyla-cetamidinate precursor along with the H2 plasma. Thedeposition process proceeded in an ideal self-limiting ALDfashion with a reasonably high saturated growth rate of 0.071nm/cycle below 100 °C. The process temperature could evenbe as low as 50 °C, so the critical agglomeration issueassociated with the Cu thin films was largely suppressed. TheCu films deposited at 50 °C were pure, continuous, smooth,and highly conformal, with the resistivity comparable to PVDCu films, demonstrating the high promise of this process formicroelectronic applications. In-situ QCM and OES measure-ments were followed to understand the reaction mechanism,and the results confirmed the high reactivity at low temperaturefor both reactants. To the best of our knowledge, this is the firstreport of implementing metal amidinate precursors for low-temperature (∼50 °C) ALD. As metal amidinates generallyshow clean surface chemistry, we believe that the strategy ofusing them in combination with H2 plasma can be largelygeneralized and extended to other metal amidinate precursorsfor depositing various high-quality metal thin films at lowtemperature.
■ ASSOCIATED CONTENT*S Supporting InformationThe Supporting Information is available free of charge on theACS Publications website at DOI: 10.1021/acs.chemma-ter.5b02137.
Thermal gravimetric analysis of the synthesized Cu(amd)precursor, SEM images of the films deposited withdifferent RF power, AFM images of the films depositedat various temperatures, and time-resolved OES results.(PDF)
■ AUTHOR INFORMATIONCorresponding Authors*E-mail: [email protected].*E-mail: [email protected].
Author ContributionsZ. G. and H. L. contributed equally to this work.
NotesThe authors declare no competing financial interest.
■ ACKNOWLEDGMENTSThis work was financially supported by NSFC (Grant Nos.51302007, 11175024, and 11375031), Shenzhen Science andTechnology Innovat ion Committee (Grant Nos .JCYJ20130329181509637 and JCYJ20140417144423201),and the Importation and Development of High-Caliber TalentsProject of Beijing Municipal Institutions (Grant Nos.CIT&TCD201404130 and KM201510015002). X.W. wouldlike to thank Professor Roy G. Gordon at Harvard Universityfor valuable discussion and support.
■ REFERENCES(1) International Technology Roadmap for Semiconductors. http://www.itrs.net/ (accessed July 2015).(2) George, S. M. Atomic Layer Deposition: An Overview. Chem.Rev. 2010, 110, 111−131.
(3) Miikkulainen, V.; Leskela, M.; Ritala, M.; Puurunen, R. L.Crystallinity of Inorganic Films Grown by Atomic Layer Deposition:Overview and General Trends. J. Appl. Phys. 2013, 113, 021301.(4) Johnson, R. W.; Hultqvist, A.; Bent, S. F. A Brief Review ofAtomic Layer Deposition: From Fundamentals to Applications. Mater.Today 2014, 17, 236−246.(5) Knisley, T. J.; Ariyasena, T. C.; Sajavaara, T.; Saly, M. J.; Winter,C. H. Low Temperature Growth of High Purity, Low ResistivityCopper Films by Atomic Layer Deposition. Chem. Mater. 2011, 23,4417−4419.(6) Gordon, P. G.; Kurek, A.; Barry, S. T. Trends in CopperPrecursor Development for CVD and ALD Applications. ECS J. SolidState Sci. Technol. 2015, 4, N3188−N3197.(7) Lee, B. H.; Hwang, J. K.; Nam, J. W.; Lee, S. U.; Kim, J. T.; Koo,S.-M.; Baunemann, A.; Fischer, R. A.; Sung, M. M. Low-TemperatureAtomic Layer Deposition of Copper Metal Thin Films: Self-LimitingSurface Reaction of Copper Dimethylamino-2-propoxide withDiethylzinc. Angew. Chem., Int. Ed. 2009, 48, 4536−4539.(8) Vidjayacoumar, B.; Emslie, D. J. H.; Clendenning, S. B.;Blackwell, J. M.; Britten, J. F.; Rheingold, A. Investigation of AlMe3,BEt3, and ZnEt2 as Co-reagents for Low-Temperature Copper MetalALD/Pulsed-CVD. Chem. Mater. 2010, 22, 4844−4853.(9) Kalutarage, L. C.; Clendenning, S. B.; Winter, C. H. Low-Temperature Atomic Layer Deposition of Copper Films Using BoraneDimethylamine as the Reducing Co-reagent. Chem. Mater. 2014, 26,3731−3738.(10) Kang, S.-W.; Yun, J.-Y.; Chang, Y. H. Growth of Cu Metal Filmsat Room Temperature Using Catalyzed Reactions. Chem. Mater. 2010,22, 1607−1609.(11) Wu, L.; Eisenbraun, E. Hydrogen Plasma-Enhanced AtomicLayer Deposition of Copper Thin Films. J. .Vac. Sci. Technol., B 2007,25, 2581−2585.(12) Niskanen, A.; Rahtu, A.; Sajavaara, T.; Arstila, K.; Ritala, M.;Leskela,̈ M. Radical-Enhanced Atomic Layer Deposition of MetallicCopper Thin Films. J. Electrochem. Soc. 2005, 152, G25−G28.(13) Jezewski, C.; Lanford, W. A.; Wiegand, C. J.; Singh, J. P.; Wang,P.-I.; Senkevich, J. J.; Lu, T.-M. Inductively Coupled HydrogenPlasma-Assisted Cu ALD on Metallic and Dielectric Surfaces. J.Electrochem. Soc. 2005, 152, C60−C64.(14) Moon, D.-Y.; Han, D.-S.; Shin, S.-Y.; Park, J.-W.; Kim, B. M.;Kim, J. H. Effects of the Substrate Temperature on the Cu Seed LayerFormed Using Atomic Layer Deposition. Thin Solid Films 2011, 519,3636−3640.(15) Profijt, H. B.; Potts, S. E.; van de Sanden, M. C. M.; Kessels, W.M. M. Plasma-Assisted Atomic Layer Deposition: Basics, Oppor-tunities, and Challenges. J. Vac. Sci. Technol., A 2011, 29, 050801.(16) Ten Eyck, G. A.; Senkevich, J. J.; Tang, F.; Liu, D.; Pimanpang,S.; Karaback, T.; Wang, G. C.; Lu, T. M.; Jezewski, C.; Lanford, W. A.Plasma-Assisted Atomic Layer Deposition of Palladium. Chem. Vap.Deposition 2005, 11, 60−66.(17) Johansson, M.; Lytken, O.; Chorkendorff, I. The StickingProbability for H2 on Some Transition Metals at a Hydrogen Pressureof 1 bar. J. Chem. Phys. 2008, 128, 034706.(18) Lim, B. S.; Rahtu, A.; Gordon, R. G. Atomic Layer Deposition ofTransition Metals. Nat. Mater. 2003, 2, 749−754.(19) Li, Z.; Rahtu, A.; Gordon, R. G. Atomic Layer Deposition ofUltrathin Copper Metal Films from a Liquid Copper(I) AmidinatePrecursor. J. Electrochem. Soc. 2006, 153, C787−C794.(20) Dai, M.; Kwon, J.; Halls, M. D.; Gordon, R. G.; Chabal, Y. J.Surface and Interface Processes During Atomic Layer Deposition ofCopper on Silicon Oxide. Langmuir 2010, 26, 3911−3917.(21) Ma, Q.; Guo, H.; Gordon, R. G.; Zaera, F. Uptake of CopperAcetamidinate ALD Precursors on Nickel Surfaces. Chem. Mater.2010, 22, 352−359.(22) Ma, Q.; Guo, H.; Gordon, R. G.; Zaera, F. Surface Chemistry ofCopper(I) Acetamidinates in Connection with Atomic LayerDeposition (ALD) Processes. Chem. Mater. 2011, 23, 3325−3334.
Chemistry of Materials Article
DOI: 10.1021/acs.chemmater.5b02137Chem. Mater. 2015, 27, 5988−5996
5995
Dow
nloa
ded
by U
NIV
TO
WN
OF
SHE
NZ
HE
N o
n Se
ptem
ber
9, 2
015
| http
://pu
bs.a
cs.o
rg
Pub
licat
ion
Dat
e (W
eb):
Aug
ust 2
0, 2
015
| doi
: 10.
1021
/acs
.che
mm
ater
.5b0
2137
(23) Ma, Q.; Zaera, F.; Gordon, R. G. Thermal Chemistry ofCopper(I)-N,N′ -di-sec-butylacetamidinate on Cu(110) Single-CrystalSurfaces. J. Vac. Sci. Technol., A 2012, 30, 01A114.(24) Li, Z.; Barry, S. T.; Gordon, R. G. Synthesis and Character-ization of Copper(I) Amidinates as Precursors for Atomic LayerDeposition (ALD) of Copper Metal. Inorg. Chem. 2005, 44, 1728−1735.(25) Krisyuk, V.; Aloui, L.; Prud’homme, N.; Sarapata, B.; Senocq, F.;Samelor, D.; Vahlas, C. CVD of Pure Copper Films from a NovelAmidinate Precursor. ECS Trans. 2009, 25, 581−586.(26) Russell, S. W.; Rafalski, S. A.; Spreitzer, R. L.; Li, J.; Moinpour,M.; Moghadam, F.; Alford, T. L. Enhanced Adhesion of Copper toDielectrics via Titanium and Chromium Additions and SacrificialReactions. Thin Solid Films 1995, 262, 154−167.(27) Yang, C.-Y.; Chen, J. S. Investigation of Copper Agglomerationat Elevated Temperatures. J. Electrochem. Soc. 2003, 150, G826−G830.(28) Barnat, E. V.; Nagakura, D.; Wang, P.-I.; Lu, T.-M. Real TimeResistivity Measurements During Sputter Deposition of UltrathinCopper Films. J. Appl. Phys. 2002, 91, 1667−1672.(29) Rocklein, M. N.; George, S. M. Temperature-Induced ApparentMass Changes Observed During Quartz Crystal MicrobalanceMeasurements of Atomic Layer Deposition. Anal. Chem. 2003, 75,4975−4982.(30) Heil, S. B. S.; van Hemmen, J. L.; van de Sanden, M. C. M.;Kessels, W. M. M. Reaction Mechanisms During Plasma-AssistedAtomic Layer Deposition of Metal Oxides: A Case Study for Al2O3. J.Appl. Phys. 2008, 103, 103302.(31) Fantz, U.; Heger, B. Spectroscopic Diagnostics of theVibrational Population in the Ground State of H2 and D2 Molecules.Plasma Phys. Controlled Fusion 1998, 40, 2023−2032.(32) Goyette, A. N.; Jameson, W. B.; Anderson, L. W.; Lawler, J. E.An Experimental Comparison of Rotational Temperature and GasKinetic Temperature in a H2 Discharge. J. Phys. D: Appl. Phys. 1996,29, 1197−1201.
Chemistry of Materials Article
DOI: 10.1021/acs.chemmater.5b02137Chem. Mater. 2015, 27, 5988−5996
5996
Dow
nloa
ded
by U
NIV
TO
WN
OF
SHE
NZ
HE
N o
n Se
ptem
ber
9, 2
015
| http
://pu
bs.a
cs.o
rg
Pub
licat
ion
Dat
e (W
eb):
Aug
ust 2
0, 2
015
| doi
: 10.
1021
/acs
.che
mm
ater
.5b0
2137
